Re: [Libguestfs] [PATCH] v2v: add support for virtio-scsi
On Wed, Apr 13, 2016 at 06:39:55PM +0100, Richard W.M. Jones wrote:
On Wed, Apr 13, 2016 at 07:45:42PM +0300, Roman Kagan wrote:
> On Wed, Apr 13, 2016 at 10:11:54AM +0100, Richard W.M. Jones wrote:
> > On Tue, Apr 12, 2016 at 06:46:31PM +0300, Roman Kagan wrote:
> > > + (* Presence of virtio-scsi controller. *)
> > > + let has_virtio_scsi =
> > > + let obj = Xml.xpath_eval_expression xpathctx
> > > +
"/domain/devices/controller[@model='virtio-scsi']" in
> >
> > I guess this short cut is OK. A true test would involve checking the
> > <target bus="scsi"> on each disk and matching it back to the
> > controller. In other words, a huge pain! Maybe you can add an
"XXX"
> > comment in the source about this.
>
> On a second look, I'm not sure I got your comment right.
>
> AFAICS there's no way in libvirt xml to define multiple scsi buses,
I think this is possible, although massively unlikely in practice. I
think the different controllers would have different
index="0"/index="1"/... attributes.
How the binding of drives to controllers is done, then? Is it specified
in drive's "address"?
And indeed we probably don't have to worry about this case.
> so all the matching I can do is just to notice the presence of
> virtio-scsi controller, and attribute all SCSI drives to it. This
> is exactly what happens in the patch.
>
> Am I missing something?
So first off I think this part of the patch and your assumption is
fine, there's no need to change the code.
However there is a possibility that the source guest might have
multiple mixed virtio-blk + virtio-scsi disks. In that case the disks
would look like this:
<disk type='file' device='disk'>
<driver name='qemu' type='raw'/>
<source file=...>
<target dev='vda' bus='virtio'/>
</disk>
<disk type='file' device='disk'>
<driver name='qemu' type='raw'/>
<source file=...>
<target dev='sda' bus='scsi'/>
</disk>
<controller type="scsi" index="0"
model="virtio-scsi"/>
It's not something that we should worry about, and a simple comment
with "XXX" in it is fine, if you agree with my analysis.
This case is handled fine in the patch, isn't it? has_virtio_scsi is a
local variable which is only consulted when populating per-drive
s_controller field: for drives with bus='scsi' it differentiates between
Source_virtio_SCSI and Source_SCSI.
The place where we make up a single disk type is
v2v.ml:rcaps_from_source, but there I explicitly error out if the guest
has multiple disk types (which may be too harsh but simplifies things
for now).
Roman.